On 6/19/24 16:29, Daniel Wagner wrote:
> On Wed, Jun 19, 2024 at 04:16:40PM GMT, Nilay Shroff wrote:
>> + # start iteration from ns-id 2 because ns-id 1 is created
>> + # by default when nvme target is setup. Also ns-id 1 is
>> + # deleted when nvme target is cleaned up.
>> + for ((i = 2; i <= iterations; i++)); do {
>> + truncate -s "${NVME_IMG_SIZE}" "$(_nvme_def_file_path).$i"
>> + _create_nvmet_ns "${def_subsysnqn}" "${i}" "$(_nvme_def_file_path).$i"
>> +
>> + # allow async request to be processed
>> + sleep 1
>
> This looks a bit fragile to ensure all request have been processed. Would
> it possible to wait on a state? E.g. something like
>
> nvmf_wait_for_state()
>
> ?
OK that's a good idea! I think it's possible to wait using _find_nvme_ns() instead
of using sleep here. We can write a wrapper around _find_nvme_ns() and wait until
namespace is ready/created.
I will send next patch with the above change.
>
>> +
>> + _remove_nvmet_ns "${def_subsysnqn}" "${i}"
>> + rm "$(_nvme_def_file_path).$i"
>> + }
>
Thanks,
--Nilay
